|
<br />1362
<br />
<br />BESTGEN ET AL.
<br />
<br />TABLE I.-Huggins models to estimate abundance (derived from model parameters), survival (S), probability of capture (P),
<br />and transition rate (1\1) among reaches for Colorado pikeminnow in the Green River basin, 2000-2003. Probability of recapture is
<br />treated as being equal to P. Covariates include river reach, sampling year or pass, and fish total length (TL). The additive effect in
<br />model 4 is a term to test for differences among capture and recapture rates; S(.) indicates a constant survival rate over all years
<br />with no covariates.
<br />
<br /> Akaike Model Number of
<br />Model AICc dAICc weight likelihood parameters Deviance
<br />1. S(TL, TL2, Te), lJi(reach, TL), p(reach X year X pass, TL, TL2) 12,926.41 0.00 0.35 1.00 81 12,760.75
<br />2. S(TL, TL2, TL'), lJi(reach, TL), p(reach X year X pass, TL, TL2, Te) 12,927.20 0.79 0.24 0.67 82 12,759.44
<br />3. S(year, TL, TL2, TL'), lJi(reach, TL), p(reach X year X pass, TL, TL2) 12,928.09 1.68 0.15 0.43 83 12,758.24
<br />4. S(TL, TL2, TL 3), lJi(reach, TL), p(reach X year X pass, TL, TL2), 12,928.47 2.06 0.12 0.36 82 12,760.71
<br />additive factor
<br />5. S(TL, TL2), lJi(reach, TL), p(reach X year X pass, TL, TL 2) 12,929.82 3.41 0.06 0,18 80 12,766.24
<br />6. S(.), lJi(reach, TL), p(reach X year X pass, TL, TL2) 12,930.25 3.84 0.05 0.15 78 12,770.86
<br />7. S(TL, TL", TL3), lJi(reach), p(reach X year X pass, TL, TL2) 12,931.82 5.42 0.02 om 80 12,768.25
<br />8. S(TL, TL2, TL'), lJi(reach, TL), p(reach X year X pass) 12,940.20 13.79 0.00 0.00 72 12,793.31
<br />9. S(.), lJi(reach), p(reach X year X pass) 12,989.48 63.07 0.00 0.00 75 12,836.34
<br />10, S(year, TL), lJi(reach, TL), p(year X pass, TL) 13,178.27 251.86 0.00 0.00 38 13,101.46
<br />11. S(year, TL), lJi(reach, TL), p(reach, year X pass, TL) 13,180.71 254.30 0.00 0.00 42 13,095.72
<br />
<br />reaches other than those normally sampled by the
<br />ISMP. Global models that fit parameters p, S, and ^ for
<br />all years and river reaches were compared with models
<br />with a reduced parameter set, and AIC c was used for
<br />model selection. We were careful to guard against
<br />overfitting the models with the sometimes sparse data
<br />available for some reaches or rivers and focused on
<br />those that gave reasonable estimates of the parameters
<br />critical to understanding the status of Colorado pike-
<br />minnow in the Green River basin.
<br />
<br />Results
<br />Fish Captures and Model Selection
<br />
<br />A total of 3,800 Colorado pikeminnow were
<br />captured during the 42 sampling passes in the five
<br />reaches from 2000 to 2003. Of those fish, 3,212
<br />(84.5%) were captured in only 1 year and not seen in
<br />any other year, 532 (14%) were captured in 2 years, 48
<br />(1.3%) were captured in 3 years, and 8 (0.2%) were
<br />captured in all 4 sampling years. We estimated the
<br />abundance of Colorado pikeminnow adults and recruits
<br />in each of the five river reaches of the Green River
<br />basin for each sampling year to determine spatial
<br />abundance patterns and temporal dynamics. Eleven
<br />models were fit to the data to examine the importance
<br />of year-specific apparent survival (S), reach transition
<br />probabilities (\)1), and capture probabilities (P) (Table
<br />1). The best model in the set had 35% of the AIC
<br />weight, the second-highest-ranked model differing onl;
<br />in the degree of the polynomial of TL used to model
<br />the capture probabilities. The third-best model with
<br />variable survival over all three annual intervals had a
<br />very high and unreliable survival estimate for 2002-
<br />2003 and was not considered. Therefore, we made
<br />inferences from the top-ranked model. That model had
<br />81 parameters and included quadratic and cubic effects
<br />
<br />of TL to model a single S over the 2000-2003 period
<br />(three length effects and S = 4 parameters), a linear
<br />effect of TL to model \)1 (a single length effect plus 20
<br />\)1 parameters, one each for fish in a given reach
<br />moving to any of the other four reaches, = 21 total
<br />parameters), and quadratic effects of TL to model p
<br />(two length effects plus those for five reaches, three
<br />sampling passes per year, and three or four sampling
<br />years [depending on the reach] = 56 parameters). We
<br />held constant the effects of length on estimated
<br />parameters, making the reasonable assumption that
<br />the effects of length on capture rate were similar
<br />riverwide and across years. The result was size-
<br />dependent probability-of-capture relationships whose
<br />magnitudes varied across passes, reaches, and years.
<br />
<br />River Reach Abundance Estimates
<br />The abundance of adult Colorado pikeminnow
<br />declined in every river reach of the Green River basin
<br />over the duration of the study, based on trends in point
<br />estimates over time (Table 2; Figure 3). Adult Colorado
<br />pikeminnow in the Yampa River reach had the lowest
<br />abundance of the five study reaches and point estimates
<br />declined 29%, from about 317 (SE = 105) fish in 2000
<br />to about 224 (SE = 75) fish in 2003. However, the 95%
<br />confidence intervals overlapped among pairs of point
<br />estimates, suggesting the estimates were not signifi-
<br />cantly different from each other. Regression of
<br />abundance as a function of time (n = 4) showed a
<br />negative relationship (loge[abundance] = 243.03 -
<br />0.1l86'year; r2 = 0.85, P = 0.079). Except for 2001
<br />(coefficient of variation [CY = 100' SDjmean] = 15%),
<br />Yampa River estimates had relatively low precision
<br />(CY = 31-34%). This was especially true in 2002 and
<br />2003, when there were no recaptures of fish captured
<br />and released in the respective years. The number of
<br />
|